Download as pdf or txt
Download as pdf or txt
You are on page 1of 27

Journal of Conflict Resolution

55(1) 106-132
ª The Author(s) 2011
Selection, Availability, Reprints and permission:
sagepub.com/journalsPermissions.nav
and Opportunity: The DOI: 10.1177/0022002710374713
http://jcr.sagepub.com

Conditional Effect of
Poverty on Terrorist
Group Participation
Jennifer Kavanagh1

Abstract
Poverty is often identified as a determinant of terrorist group participation, but existing
research reveals mixed support for this relationship. Some studies find that macroeco-
nomic decline is associated with increased production of terrorists, but micro-level
research suggests terrorists have above average socioeconomic status and educational
attainment. In this article, the author argues that poverty should increase terrorist
group participation only for individuals with high education. The author suggests that
as a result of terrorist group selection preferences and the lower opportunity costs for
militant group membership in economically depressed environments, the likelihood of
terrorist group participation should be highest for the highly educated, poor members
of any population. The author tests the hypotheses using data from Krueger and
Maleckova (2003) on participation in Hezbollah, adding an interaction term to their
model. The results support the hypotheses. Poverty increases the likelihood of partic-
ipation in Hezbollah only for those with at least high school education.

Keywords
education, poverty, and terrorism, counterterrorism, opportunity costs, quality of
terrorism

1
Department of Political Science and Gerald R. Ford School of Public Policy, University of Michigan,
Ann Arbor, MI, USA

Corresponding Author:
Jennifer Kavanagh, University of Michigan, 735 South State Street, Ann Arbor, MI 48109, USA
Email: jennkav@umich.edu

106
Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015
Kavanagh 107

Introduction
As the threats from transnational and domestic terrorism become increasingly wide-
spread, there is an intensified interest in identifying country and individual-level
characteristics that catalyze the development of violent organizations and the per-
sonal decision to join a terrorist group (Enders and Sandler 2000, 2006). Existing
literature discusses political, economic, social, and even psychological factors at
both the macro- and micro-levels that are hypothesized to promote or impede the
formation and growth of terrorist organizations. Economic arguments receive an
especially large amount of attention in both the academic and the popular press. The
simplest economic argument blames poverty and unemployment for creating des-
peration and hopelessness among the masses, which drives them to join terrorist
organizations, either to receive the social services such groups provide or due to a
general lack of other opportunities. Research that tests for this relationship finds a
more complicated set of patterns. Specifically, while macroeconomic decline does
seem to be associated with higher incidence of terrorism, terrorists themselves are
not drawn from the poor, uneducated elements of society.
This article integrates these two sets of findings by identifying an alternative
mechanism through which worsening economic conditions affect the nature and
intensity of terrorism, one that does not rely on increased participation by poor,
uneducated individuals. I propose a conditional relationship between poverty, educa-
tion, and terrorist group involvement and argue that poverty increases the likelihood
of membership in terrorist organizations only for those with high education. I suggest
that as a result of terrorist group selection preferences and the low opportunity costs
for militant group membership that exist in economically depressed macroeconomic
environments, highly educated, poor individuals in any society should be most likely
to participate in terrorist group activities. Using data from Krueger and Maleckova
(2003), I test these hypotheses specifically with reference to Lebanon and participa-
tion in Hezbollah. I find that while poverty on its own decreases the probability of
participation in Hezbollah, poverty does appear to be a significant positive predictor
of participation in Hezbollah for highly educated individuals.1 I conclude with a dis-
cussion of the findings and their policy implications.

Existing Literature
Literature focused on the production of terrorism falls into two categories. Macro-
level studies identify country characteristics that foster and impede the formation
of terrorist groups and the production of terrorists who perpetrate violent acts at home
and abroad. Micro-level work addresses the individual attributes associated with the
decision to join and participate in terrorist groups. While there is some overlap across
macro- and micro-level research, the two approaches ask rather different questions.
Literature focused on the national level production of terrorist organizations iden-
tifies economic, political, and demographic characteristics that encourage or support

107
Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015
108 Journal of Conflict Resolution 55(1)

terrorist group formation and vitality. For instance, there is some evidence that the
existence of ungovernable areas or a large population of recruitable young men pro-
motes the successful formation of terrorist organizations (Homer-Dixon 2002).
Weak political and civil rights, government instability, high levels of repression, and
weak political institutions are also consistently associated with increased likelihood
of terrorist production (Krueger and Maleckova 2003; Krueger 2007; White 1989,
1993; Lichbach 1987; Krueger and Laitin 2008). Findings on economic predictors
of terrorist group formation are more mixed. Some work argues that political char-
acteristics such as institutional strength and political freedom are stronger predictors
of terrorist group formation than macroeconomic variables (Krueger and Laitin
2008). Other work takes the opposite approach, proposing that national-level eco-
nomic factors are significantly associated with terrorist production. This literature
finds strong relationships between low or falling national income, high rates of
unemployment, low economic openness, and the production of terrorists (Blomberg,
Hess, and Weerepana 2004; Blomberg and Hess 2008a; Blomberg and Hess 2008b;
Honaker 2008). A final set of studies shows evidence for the relevance of both
strong government institutions and economic factors as predictors of terrorist pro-
duction. Specifically, fewer terrorists originate from countries with more robust
government-sponsored social welfare programs (Burgoon 2006).
Literature on the micro-level production of terrorists explores individual charac-
teristics that shape the decision to join terrorist groups. Some studies consider emo-
tional and psychological attributes. Although early research on terrorism suggested
that terrorists suffered from social deviance and mental illness, more recent work has
rejected such arguments (Hudson 1999; Sageman 2004). Other explanations focus
on personal and emotional experiences. For example, interviews with terrorist group
members identify the loss of a family member to counterterror operations or the wit-
nessing of a violation of ‘sacred values’ as common reasons individuals choose to
join terrorist organizations (Atran 2006; Brym and Araj 2005). Sociological factors
such as small group dynamics, pressure for conformity, and the search for belonging
can also catalyze radicalization (Sageman 2004; Sprinzak 1990). There is evidence
that political freedom and access, considered above at the country level, also affect
individual participation decisions. Violent resistance may become more appealing
when political channels are blocked (Crenshaw 2001). Finally, attributes such as
personal values, ideological commitment, and the expected psychological and mate-
rial benefits of participation directly shape individual choices about terrorist group
participation. Existing research on the relationships between individual economic
status, educational attainment, and terrorist group participation consistently finds
that low education and poverty are poor predictors of terrorist group participation
or support (Sageman 2004). In fact, there is considerable evidence that individuals
with higher levels of education are more likely to support the use of terrorist vio-
lence, suggesting that radicalization may be facilitated by some base level of educa-
tion (Krueger and Maleckova 2003; Angrist, 1995). Studies of Hezbollah fighters
and Palestinian suicide bombers have independently confirmed that members of

108
Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015
Kavanagh 109

these groups tend to have higher economic status and education levels than the
referent populations and even than other members of terrorist groups, who occupy
less central roles (Krueger and Maleckova 2003; Berrebi 2003; Hassan 2001;
Benmelech and Berrebi 2007).
The existing literature suggests a range of factors that influence formation of ter-
rorist organizations and the individual decision to join such groups but comes to no
real consensus on the relationship between poverty and terrorism. Macro-level
research predicts some association between macroeconomic weakness and increased
production of terrorists, but micro-level analysis finds that poor, uneducated individ-
uals are not more likely to be members of these groups. This article addresses the
tension and inconsistency in existing literature, proposing a conditional relationship
between poverty, education, and participation that integrates individual-level out-
comes with macro-level processes.

Argument
In this section, I suggest the existence of a conditional relationship between poverty,
education, and terrorist group membership, which makes highly educated, poor indi-
viduals the most likely members of terrorist organizations. I propose two theoretical
mechanisms that contribute to this outcome, terrorist group selection preferences
and relative opportunity costs to terrorist group membership.
First, because terrorist groups focus their recruiting resources on individuals they
expect to become the most productive terrorists, they are likely to expend especially
strong efforts to attract highly educated and skilled individuals. Existing models and
analysis of terrorist group activities suggest that individuals with higher levels of
education are more productive workers in the market economy and more efficient
terrorists. For instance, studies of Palestinian suicide bombers show that missions
executed by bombers with high levels of education or high-skill occupations are
more likely to be successful and to result in high numbers of Israeli casualties. Ter-
rorist groups appear to recognize the productivity gains provided by these highly
educated individuals, strategically employing their most highly qualified operatives
on high-profile missions with potentially significant impacts (Benmelech and Ber-
rebi 2007). Bueno de Mesquita (2005) formalizes these selection effects in a model
that describes the effects of macroeconomic destruction caused by counterterrorism
on the quality of terrorist violence. In the model, counterterror actions destroy the
target economy and eventually eliminate opportunities for even highly skilled work-
ers. He argues that terrorist groups focus recruiting efforts on these newly unem-
ployed, high-skill individuals for the additional productivity they provide. He uses
the model to show that macroeconomic decline, by raising the quality of unem-
ployed, potential terrorist group recruits, may also indirectly increase the ‘‘quality’’
of terrorism perpetrated by the organization. The same may be true under conditions
of long-term economic depression, where market opportunities have been elimi-
nated over time, gradually increasing the quality of the pool of potential recruits.

109
Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015
110 Journal of Conflict Resolution 55(1)

The second determinant of the conditional relationship between terrorist group


participation, education, and poverty concerns relative opportunity costs to partici-
pation in militant activities. In a strong market economy, more highly educated and
skilled individuals have lucrative market options and high opportunity cost to par-
ticipation in terrorist group operations. However, in weak or impoverished econo-
mies, the lack of market options reduces the opportunity costs of group
participation for highly educated individuals. In contrast, opportunity costs to parti-
cipations for less educated individuals are likely to be fairly constant across eco-
nomic conditions. Lower opportunity costs in weak economic markets make these
educated individuals more susceptible and available for recruitment, increasing the
likelihood of their membership (Bueno de Mesquita 2005; Berman and Laitin 2006).
The combination of selection preferences and reduced opportunity costs create a
situation where highly educated, poor individuals are both more available and more
heavily targeted for recruiting than individuals who are not poor or those who are
poor, but uneducated. This inference has two implications. First, following eco-
nomic shocks that reduce macroeconomic opportunities, the proportion of highly
educated terrorist recruits will rise even as the size of the group stays the same. Sec-
ond, in any given population, highly educated, poor individuals should be most
likely to be active members of terrorist groups. Individuals who are poor and have
little education, however, should be no more likely to participate than average. Pov-
erty status, therefore, should predict participation in terrorist activities only for the
most highly educated individuals in a given society. This article focuses on the sec-
ond, more static implication:

Hypothesis 1: Poverty will predict participation in terrorist groups for highly edu-
cated individuals but not those with low education.
Hypothesis 2: Participation in militant group activity should be highest among
individuals who are both highly educated and poor.

Data
I test my hypotheses with specific reference to the case of Hezbollah using data from
and extending the analysis presented in Krueger and Maleckova (2003). The data
used for the article combines demographic information on 129 deceased Hezbollah
militants, who died in action from 1982 to 1994 with the 1996 Lebanese Population
and Housing Survey (PHS). The PHS sample is restricted to include only those indi-
viduals aged 15 to 38, the range covered by the Hezbollah militant sample, resulting
in 120,796 observations. According to Krueger and Maleckova, two-thirds of the
Hezbollah militants were killed between 1986 and 1988. Biographical information
on the militants was extracted from Hezbollah’s internal newspaper, Al-Ahd, by Eli
Hurvitz. Krueger and Maleckova coded these bibliographies to capture the demo-
graphic variables most comparable to those recorded in the PHS. The key variables
include education level, poverty, age, location of residence, and membership in

110
Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015
Kavanagh 111

Hezbollah. The ‘‘poverty’’ variable is coded [0,1] based on a set of living condition
indicators, such as number of rooms per person, access to running water, occupation,
and car ownership. The use of these indicators helps to remove some of the subjec-
tivity associated with the self-reported poverty measure. The education variable
codes the respondent’s highest completed level of education: primary, preparatory,
high school, university, and post-university. Krueger and Maleckova, however, use
an aggregated form of this variable, which takes a value of ‘‘1’’ if the individual has
completed high school or greater and a value of ‘‘0’’ otherwise. The relevant depen-
dent variable, participation in Hezbollah, is also dichotomous, and heavily skewed
toward 0.2 I include descriptive statistics for the relevant variables in table 1, and
by subgroup, in table 2.
The decision to merge the 1996 PHS with the data on the Hezbollah militants is
justified by similarities between the Hezbollah participants and the general popula-
tion sample taken from the PHS but relies on several assumptions. Tables 1–3 show
close similarities between the average age, education, and poverty levels of Hezbol-
lah militants and the general population surveyed in the PHS. The two samples seem
broadly comparable, despite the difference in sample size and time period of the data
collection. Most significantly, despite the generally more depressed economic condi-
tions that prevailed in Lebanon during the 1980s, Hezbollah members were no more
likely than members of the general population or the Shia subsample the criteria for
‘poverty,’ as defined previously. Hezbollah militants are also more likely to have high
levels of education. Just these basic descriptive statistics offer some initial support for
the hypotheses above, since the most significant difference between the Hezbollah
sample and the general population appears to not be poverty status, but education.
However, while these similarities support the decision to pool the two samples,
the synthesis requires several assumptions. First, it assumes that there no are signif-
icant time-varying covariates that make the PHS and Hezbollah militant populations
fundamentally different. Most importantly for tests of Hypothesis 1, merging the
data assumes general macroeconomic stability across periods. Significant changes
in macroeconomic opportunities would bias a comparison of the two populations.
Fortunately, characteristics of Lebanese macroeconomic development during the
1980s and 1990s suggest that this assumption may be a justifiable one. The Lebanese
economy was severely depressed in the 1980s as a result of a long running civil war,
the period during which most Hezbollah fighters were killed, and slowly began to
recover only after 1993. Even then, however, ‘‘lack of adequate infrastructure, a
weak institutional framework, and shortages of human capital continued to constrain
overall economic recovery’’ (Eken et al. 1995, 1). In addition, what economic prog-
ress did occur was concentrated among the country’s elite, with few positive impli-
cations for the economic well-being of the poorer Shia communities in Southern
Lebanon from which most Hezbollah militants were drawn. One report on Leba-
non’s postwar recovery noted in the late 1990s that although ‘‘the country’s econ-
omy has recovered during the past decade, this has not benefited the poor’’ living
in ‘‘the ‘misery belt’ of southern Beirut’’ (Knudsen 2005, 3). As a result, although

111
Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015
112 Journal of Conflict Resolution 55(1)

Table 1. Descriptive Statistics, Full Sample

Standard
Variable N Mean Error Min Max

Age 120,925 25.564 0.020 15 38


Education 120,796 2.919 0.005 1 5
Hezbollah 120,796 0.001 0.000 0 1
Poverty 120,874 0.326 0.001 0 1
South 120,796 0.095 0.001 0 1
Beirut 120,925 0.128 0.001 0 1
>HS educ. 120,846 0.379 0.001 0 1
Shia 120,925 0.161 0.001 0 1
Weight 120,877 10.757 0.001 3.66 10.76
term
Pov  Ed 120,925 0.034 0.001 0 1

Lebanon’s macroeconomic conditions in the 1980s were certainly worse than those
in the 1990s, the economic well-being of average Lebanese, including the majority
of those polled in the PHS, remained largely similar. Although the pooled data set
is not ideal, the broad similarity in external conditions appears sufficient to
justify the merging of the two populations and the use of this combined sample to
explore the conditional relationship between poverty, education, and terrorist group
participation.
Several additional assumptions are made in the decision to merge the two popu-
lations. First, this synthesis assumes that the information on Hezbollah fighters pre-
sented in Al-Ahm is both accurate and comparable to that found in the PHS. The
discussion provided by Krueger and Maleckova (2003) implies that this condition
is met but suggests several caveats. First, because the Hezbollah militants in the sam-
ple are martyrs, involved in relatively high-profile operations, they may not be rep-
resentative of the majority of Hezbollah members. In addition, even within this set of
Hezbollah martyrs there is a range of different activities, some more violent than
others, but all coded simply as ‘‘participation.’’ The data do not capture non-
martyr membership in the organization. As a result, the analysis reveals only those
covariates that separate Hezbollah martyrs from the population. Finally, the merging
of the two data sets assumes no change in the rate of participation in Hezbollah over
the 1982–1996 period. Based on Hezbollah’s continued military and electoral
strength, this seems a reasonable assumption.
Concerns about data reliability may also limit the certainty of findings. First, for
some Hezbollah militants, data were missing and had to be imputed. For example,
Krueger and Maleckova imputed poverty status based on information on the individ-
ual’s family background or using the mean value of the non-missing observations.
Second, much of the data, including poverty, rely on self-reports from the militant
group or from Lebanese civilians. The use of self-reporting creates some concerns

112
Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015
Kavanagh 113

Table 2. Descriptive Statistics, Hezbollah Sample, PHS Sample, and Shia Population (from
PHS)

Standard
Variable Mean Error Min Max

Hezbollah sample, N¼129


Age 22.183 3.169 15 38
Education* 3.551 2.055 1 7
Hezbollah 1 0 1 1
Poverty 0.249 0.282 0 1
South 0.302 0.437 0 1
Beirut 0.422 0.470 0 1
>HS educ. 0.474 0.390 0 1
Shia 1 0 1 1
Weight term 3.663 0 3.663 3.663
Pov  Ed 0.098 0.170 0 1
PHS sample N¼120,796
Age 25.567 6.787 15 38
Education 2.918 1.754 1 7
Hezbollah 0 0 0 0
Poverty 0.326 0.469 0 1
South 0.095 0.293 0 1
Beirut 0.127 0.333 0 1
>HS educ. 0.379 0.485 0 1
Shia 0.160 0.367 0 1
Weight term 10.764 0 10.764 10.764
Pov  Ed 0.034 0.182 0 1
Shia population (from PHS), N¼19,453
Age 25.18 6.69 15 38
Education** 2.88 1.95 1 7
Hezbollah 0.01 0.08 0 1
Poverty 0.48 0.50 0 1
South 0.26 0.44 0 1
Beirut 0.00 0.05 0 1
>HS educ. 0.26 0.44 0 1
Shia 1 0 1 1
Weight term 10.72 0.58 3.66 10.76
Pov  Ed 0.05 0.22 0 1

* N=78; ** N=19,402.

about the endogeneity of poverty and may bias the results. However, while it may be
difficult using this self-reported data to draw a causal inference, the argument here
suggests simply that high education and poverty increase the likelihood that an indi-
vidual will turn to terrorism. It does not exclude the possible contribution of other fac-
tors. In addition, any bias caused by self-reporting should not affect the comparison of
the two subpopulations, since it should bias both populations in similar ways.

113
Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015
114 Journal of Conflict Resolution 55(1)

Table 3. Relative Demographic Characteristics

Percentage of Percentage of Percentage


Deceased Lebanese Shia
Hezbollah Militants Population Population

Percentage of population 1 100 19


Impoverished 28 33 48
background
Education level
Illiterate 0 6 6
Read and write 22 7 11
Primary 17 23 27
Preparatory 14 26 30
Secondary 33 23 19
University 13 14 7
Postgraduate 1 1 1
Age
15–17 2 15 16
18–20 41 14 14
21–25 42 23 23
26–30 10 20 20
31–38 5 28 26
Region of residence
Beirut 42 13 .25
Mount Lebanon 0 36 0
Bekaa 26 13 41
Nabatieh 2 6 33
South 30 10 26
North 0 22 0
Source: Krueger 2007, p. 37 (first two columns). Author’s own tabulations (third column).

Although limitations of the data set reduce the certainty of any findings, the
Lebanese example is still a particularly good test case for the hypotheses pre-
sented here for several reasons. First, the long-lasting economic depression during
the Lebanese civil war created just the sort of weak market conditions in which
highly educated, poor individuals should be most susceptible to recruitment by
terrorist organizations. Second, Hezbollah itself had deep roots in Lebanese soci-
ety, widespread popularity, and sufficient resources to be particular and selective
about the individuals it recruited for militant operations (Norton 2009). As a
result, the Lebanese case is characterized by a highly selective terrorist organiza-
tion and generally weak market opportunities, both conditions that, according to
the argument here, should support a strong conditional relationship between pov-
erty, educational attainment, and terrorist group participation.

114
Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015
Kavanagh 115

Method
In their original analysis, Krueger and Maleckova explore poverty and education
as predictors of participation in Hezbollah. They apply a logit model in which the
dependent variable is Hezbollah membership and independent variables include
age, education, poverty status, and location of residence. They find that neither
poverty nor low education increase the probability of participation along with
some evidence that higher levels of education increase the likelihood of Hezbol-
lah membership.3 To test my conditional hypothesis, I add an interaction term
formed by the product of the dichotomous education and poverty variables that
will allow me to consider how the effect of poverty on participation in Hezbollah
differs for those with high and low education. If there is evidence for the hypoth-
eses, poverty will be a significant positive predictor of participation only for those
individuals with at least high school education, and the probability of participa-
tion in terrorist group activities should be highest for the highly educated, but
poor subsample of the population. I interpret the results by computing the pre-
dicted probabilities of participation for high- and low-education and poverty
groups.4 The inclusion of the interaction term makes the total effect of poverty
on Hezbollah membership dependent on education level. This means that the rela-
tionship between poverty and participation suggested by the model will be differ-
ent for those with high and low education.5 The models are estimated using the
full sample and a restricted sample that includes only the Shia referent group.
The use of the logit model is appropriate in this case despite the heavily skewed
data. First, logit models perform more effectively than probit specifications when bin-
ary choice data are heavily biased toward zero (Gujarati, 2004). Second, goodness-of-
fit tests on each specification show significant chi-square scores, indicating that the
model does closely approximate the true distribution. Finally, additional diagnostic
tests show a successful prediction rate of 99 percent using the logit specification that
includes the interaction term. I explored the use of more complex binary choice mod-
els and more flexible specifications such as generalized linear models. However, I
found that they did not provide, more closely fit the data, or change the substantive
results. As a result, I retain the logit specification chosen by Krueger and Maleckova.
One concern with this approach relates to the nature of the dependent variable,
membership in Hezbollah. The Hezbollah militants were selected into the sample
because of their participation in Hezbollah, creating the problem of choice-based
sampling. Because the sample is not random, the estimates may not be consistent
(Manski and Lerner 1977). To solve this problem, the authors create weights for the
Hezbollah and non-Hezbollah observations based on the estimated frequency of
each group in the population relative to its frequency in the sample and argue that
this should produce consistent estimates. In my analysis, I will similarly include esti-
mates from both weighted and unweighted models.
Another methodological concern relates to the use of indicator variables.
Although indicator variables simplify the analysis, aggregation may also cause us

115
Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015
116 Journal of Conflict Resolution 55(1)

to ignore more nuanced effects of changes in education or poverty that would be


apparent through the use of continuous variables. To address this challenge, I also
report the results from a model that includes the full set of education levels, prepara-
tory through higher than university education with primary education as the omitted
category. This disaggregated model shows how the marginal effect of changes in
poverty status varies in size and significance across education levels.

Results
The results support the hypotheses and show evidence of the expected conditional
relationship. Once the interaction term is included, poverty reduces the likelihood
of participation in terrorism for those with low education but increases the likelihood
of membership for those with high education. In each case, the poor, highly educated
subsample is the most likely to participate in Hezbollah’s militant activities. For all
specifications, the log-ratio test suggests that the inclusion of the interaction term
improves the overall fit of the model. Table 4 shows the full regression results
including the total effect of poverty and education when the interactive term is taken
into account. Tables 5 and 6 show the marginal effects associated with these regres-
sion results. I will focus the discussion on table 6, which compare the marginal effect
of the change from nonpoverty to poverty for high and low education. For individ-
uals who have low education, poverty has a negative marginal effect on participa-
tion. For individuals with high education, however, poverty has a significant
positive marginal effect on the probability of participation and increases the likeli-
hood of involvement in Hezbollah’s activities. Similarly, the education variable has
a positive marginal effect on participation for only poor individuals. Although the
marginal effects of changes in education and poverty status are small, because the
baseline probability of joining Hezbollah is also very small, the marginal changes
may still be substantively significant. In model 2, for example, for the high education
group, a change in poverty status increases the predicted probability of participation
from .0005 to .001, a 100 percent increase in the likelihood of participation. The
effect of the same change is smaller for model 4 but still leads to a 50 percent
increase in the predicted probability. When education is low, the change from non-
poor to poor reduces the predicted probability of participation by 50 percent in
model 4 and 80 percent in model 2.
The Shia-specific regressions offer insight into the determinants of participation in
militant activities within Hezbollahs’s Shia referent group. Several observations are
relevant. First, because all Beirut Shia are members of Hezbollah, the Beirut indicator
drops out in model 6 due to collinearity. Second, although poverty has a negative
effect on participation when education is low, it has no real effect on the participation
of the highly educated individuals. Although this result diverges slightly from those of
previous specifications, it is not inconsistent with the hypothesis, however, since
compared to their low-educated peers, the poor, highly educated Shia individuals are
more likely to participate in terrorism. Within the Shia community, then, education is

116
Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015
Table 4. Regression Results, Interaction Term Included

Weighted,
Unweighted Weighted Shia ¼ 1

1 2 3 4 5 6

Poverty 0.857 (.18)** 0.696 (.18)** 0.857 (.18)** 0.696 (.18)** 1.33 (.18)** 1.1 (.25)**
Age 0.082 (.008)** 0.082 (.008)** 0.082 (.008)** 0.082 (.008)** 0.08 (.009)** 0.07 (.01)**
Education (>HS) 0.052 (.15) 0.174 (.15) 0.052 (.15) 0.173 (.15) 0.35 (.15)* 0.37 (.22)*
Poverty  Education 1.671 (.24)** 1.72 (.24)** 1.67 (.24)** 1.717 (.23)** 1.28 (.24)** 0.99 (.32)**
^^
Beirut 2.215 (.21)** 2.215 (.21)**
South 2.175 (.22)** 2.174 (.22)** 1.09 (.24)**
Constant 4.736 (.22)** 5.765 (.24)** 5.815 (.22)** 6.846 (.24)** 3.98 (.24)** 5.344 (.37)**
R2 .03 .1 .024 .09 .05 .05
Observations 120,925 120,925 120,925 120,925 19,453 19,391
Poverty þ Pov  Ed 0.81 (.17)** 1.02 (.18)** 0.81 (.17)** 1.02 (.18)** 0.06 (.17) 0.07 (.23)
Education þ Pov  Ed 1.62 (.23)** 1.55 (.23)** 1.62 (.23)** 1.55 (.23)** 1.63 (.23)** 1.36 (.32)**
LR test vs. p < .000** p < .000** p < .000** p < .000** p < .000** p < .000**
noninteraction
model (chi-square)

Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015


Note: Robust standard errors.
*significant at .05 percent.
**significant at .01 percent.
^^ in the weighted model when the population is restricted to Shia only, the Beirut term drops out.

117
118
Table 5. Marginal Effects of Change in Poverty Status, Difference Approach

Unweighted Weighted Weighted, Shia¼1

1 2 3 4 5 6

Poverty 0.0007 (.00015)** 0.0003 (.0001)** 0.00025 (.00005)** 0.00012 (.00003)** 0.002 (.00027)** 0.0008 (.00019)**
Age 0.0001 (.00001)** 0.0000 (.00001)** 0.00002 (.000)** 0.00001 (.000)** 0.001 (.00002)** 0.00005 (.00001)**
**
Education (>HS) 0.0000 (.00012) 0.0001 (.0000) 0.00002 (.00004) 0.00003 (.00003) 0.00055 (.00023) 0.0003 (.0002)
Pov  Ed 0.0014 (.00022)** 0.0009 (.00015)** 0.00048 (.00007)** 0.00029 (.00005)** 0.002 (.00037)** 0.0008 (.00025)**
Beirut 0.0011 (.00012)** 0.00038 (.00004)**
South 0.0011 (.00012)** 0.00037 (.00004)** 0.0012 (.0003)**
Predicted .0008 .0005 .0003 .0002 .002 .0008
probability
Note: Robust standard errors.
**significant at .01 percent.

Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015


Table 6. Marginal Effects, Selected Specifications

Unweighted Weighted Weighted, Shia

Model 2 Model 4 Model 6

Change in Poverty when pr(H) ¼ 0.0005 (.00) pr(H) ¼ .0005 pr(H) ¼ 0.0001 (.00) pr(H) ¼ .0002 pr(H) ¼ 0.000 (.00) pr(H) ¼ .0001
education is high
Change in Poverty when pr(H) ¼ 0.0004 (.00) pr(H) ¼ .0005 pr(H) ¼ 0.0001 (.00) pr(H) ¼ .0002 pr(H) ¼ 0.0008 (.00) pr(H) ¼ .0007
education is low
Change in Education when pr(H) ¼ 0.0005 (.00) pr(H) ¼ .0003 pr(H) ¼ 0.0002 (.00) pr(H) ¼ .0001 pr(H) ¼ 0.0005 (.00) pr(H) ¼ .001
poverty is high
Change in Education when pr(H) ¼ 0.0001 (.27) pr(H) ¼ .0006 pr(H) ¼ 0.00004 (.27) pr(H) ¼ .0002 pr(H) ¼ 0.0005 (.06) pr(H) ¼ .0005
poverty is low

Note: p Values in parentheses for test that marginal effect is equal to zero. pr(H) ¼ baseline probability of participation.

Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015


119
120 Journal of Conflict Resolution 55(1)

0.003
P = 1, HS = 1
P = 1, HS = 0
0.0025 P = 0, HS = 1
P = 0, HS = 0

0.002

0.0015

0.001

0.0005

0
Model 2 Model 4 Model 6

Figure 1. Predicted probability of participation by subgroup.

Table 7. Predicted Probabilities of Participation by Group

Unweighted Weighted Weighted, Shia

Model 2 Model 4 Model 6

P ¼ 1, HS ¼ 1 0.003 (.002) 0.001 (.001) 0.0017 (.001)


P ¼ 0, HS ¼ 0 0.0012 (.0008) 0.0004 (.0007) 0.0001 (.0008)
P ¼ 1, HS ¼ 0 0.0005 (.002) 0.0002 (.0003) 0.0016 (.0003)
P ¼ 0, HS ¼ 1 0.0012 (.004) 0.0004 (.0005) 0.0004 (.001)
Note: Standard errors in parentheses. Baseline model holds all covariates at the mean. P ¼ 1 indicates
high poverty and HS ¼ 1 indicates high education.

the strongest predictor of participation, regardless of poverty status. The significant


high school education indicator confirms this interpretation.
A comparison of the predicted probabilities for individuals in each of the four
possible education/poverty categories shows that, as predicted, highly educated,
poor individuals are the most likely to participate in militant activities. Figure 1 and
table 7 show that the highly educated, poor individuals are the most likely to be

120
Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015
Kavanagh 121

members of Hezbollah in each specification. For models 2 and 4, a highly educated,


poor individual is about 2.5 times as likely to be a member of the terrorist group
compared to a nonpoor individual with either high or low education and at least five
times as likely to be a participant than an individual with high poverty and low edu-
cation. If we look within education groups only, these differences are even more
striking. When education is low, poor individuals are about half as likely to join the
terrorist group as nonpoor individuals. However, when education is high, poor indi-
viduals are more than twice as likely to be members of the group. We can consider
the effects of education in the same way. When poverty is high, high education
increases the likelihood of participation by at least five times. When poverty is low,
however, changes in education do not significantly affect participation. The high
poverty-high education group is also the most likely to join the terrorist organization
in the Shia-restricted sample (though its advantage over the nonpoor–high education
subgroup is not statistically significant). In this final specification, the high poverty–
high education individuals are about twice as likely as the nonpoor, low education
group to join and about 4.5 times as likely as the high poverty–low education
subgroup.
Several of the control variables are also significant across specifications. Liv-
ing in Beirut or one of the Southern Shia districts is associated with participation
in Hezbollah. Age shows a small negative relationship, suggesting the older indi-
viduals are slightly less likely to participate in the organization. Finally, the mod-
els that include the location indicators better fit the data than those omitting these
variable.

Alternative Specification
I noted above that the aggregation of the education variable into a single indicator
ignores more nuanced relationships between poverty and participation by education
level. In fact, the theory suggests that the effects of poverty on participation should
increase with each additional education level, gaining significance only at higher
thresholds. To test this, I run a separate set of specifications, including four of the
education levels, omitting primary education, and including the interactions of each
with poverty. The results show the effect of poverty on the probability of participa-
tion at each level of education as compared to the omitted, primary education refer-
ent group. I present these regression results in table 8. In table 9, I present the first
difference marginal effects for three of the six specifications, including the total
marginal effect of the poverty variable at each level of education, when the condi-
tional relationship is incorporated. Importantly, there are no poor, members of Hez-
bollah that fall in the ‘‘higher than university’’ education category in the sample, so
these results apply only through university education. The lack of observations in the
poor, higher than university education category suggests either that those individuals
with the highest education are less likely to join Hezbollah or that they do not par-
ticipate in martyrdom operations, but act as leaders or organizers.

121
Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015
122
Table 8. Regression Results, Disaggregated Education Level

Model 1A Model 2A Model 3A Model 4A Model 5A Model 6A


** ** ** ** **
Age 0.10 (.01) 0.11 (.01) 0.10 (.01) 0.11 (.01) 0.10 (.01) 0.09 (.01)**
Poverty 1.59 (.16)** 1.43 (.16)** 1.59 (.16)** 1.43 (.16)** 1.98 (.16)** 1.72 (.24)**
Preparatory 2.60 (.38)** 2.71 (.38)** 2.60 (.38)** 2.71 (.38)** 2.58 (.38)** 2.33 (.49)**
Prep  Poverty 1.69 (.56)** 1.69 (.56)** 1.69 (.56)** 1.69 (.56)** 1.43 (.56)** 1.23 (.74)
Secondary 1.65 (.25)** 1.82 (.25)** 1.65 (.25)** 1.82 (.25)** 1.38 (.25)** 1.14 (.33)**
Secondary  Pov 2.20 (.36)** 2.25 (.37)** 2.20 (.36)** 2.24 (.38)** 1.83 (.37)** 1.40 (.47)**
University 1.88 (.34)** 2.16 (.35)** 1.88 (.34)** 2.15 (.35)** 1.14 (.34)** 1.05 (.48)**
Univ  Poverty 2.86 (.34)** 3.01 (.34)* 2.85 (.34)** 3.00 (.35)** 2.29 (.36)** 2.07 (.5)**
Higher Studies 1.09 (1.0) 1.52 (1.0) 1.09 (1.0) 1.52 (1.0) 0.67 (1.0)
Beirut ¼ 1 2.41 (.21)** 2.41 (.21)**
South ¼ 1 2.16 (.22)** 2.16 (.22)** 1.18 (.25)**
** **
Constant 3.10 (.26) 4.08 (.27)** 4.18 (.26) 5.17 (.27)** 2.270.27 3.85 (.4)**
p(chi-square) > 0 .00 .00 .00 .00 0 .00
R2 .07 .15 .06 .13 .08 .08
Note: Robust standard errors.

Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015


*significant at .05 percent.
**significant at .01 percent.
Kavanagh 123

Table 9. Marginal Effects, Disaggregated Education Level

Unweighted Weighted Weighted, Shia

Model 2A Model 4A Model 6A

Poverty 0.0005 (.00) 0.0002 (.00) 0.0011 (.00)


Age 0.0000 (.00) 0.0000 (.00) 0.0001 (.00)
Poverty  University 0.0011 (.00) 0.0004 (.00) 0.0014 (.00)
Poverty  Secondary 0.0008 (.00) 0.0003 (.01) 0.0009 (.11)
Poverty  Preparatory 0.0006 (.00) 0.0002 (.00) 0.0008 (.01)
Higher studies 0.0003 (.00) 0.0001 (.00)
University 0.0004 (.00) 0.0001 (.00) 0.0005 (.00)
Secondary 0.0004 (.00) 0.0002 (.00) 0.0006 (.00)
Preparatory 0.0007 (.00) 0.0002 (.00) 0.0012 (.00)
Beirut 0.0008 (.00) 0.0003 (.00)
South 0.0008 (.00) 0.0003 (.00) 0.0011 (.00)
Predicted probability .0004 .0001 .0007
Total marginal effects
Poverty þ Poverty  Univ 0.0006* 0.0002* 0.0002*
Poverty þ Poverty  Secondary 0.0003* 0.0001* 0.0002
Poverty þ Poverty  Preparatory 0.0001 0.0000 0.0003
University þ Poverty  Univ 0.0006* 0.0002* 0.0010*
Secondary þ Poverty  Secondary 0.0003* 0.0001* 0.0004
Preparatory þ Poverty  Preparatory 0.0001 0.0000 0.0004
Note: p Values in parentheses for test that marginal effect equals zero. Results are in comparison to
omitted referent category.
1
*significant at .05 level.

The results offer further support for the hypotheses. In each case, poverty increase
the likelihood of participation only for individuals with high education, and the like-
lihood of participation is greatest among those who are poor and highly educated. As
a result, likelihood of participation is greatest among those who are poor and highly
educated. Comparing the marginal effects of the poverty variable across levels of
education, for each of the specifications in table 9, it is clear that the total effect
of poverty on the likelihood of participation rises as we move from preparatory to
secondary to university education. However, the conditional relationship is only sig-
nificant for the university and secondary education categories. This is true for each
of the specifications presented. Importantly, there is some evidence of a positive
effect of poverty on participation for university educated Shia (in model 6), a finding
not observed in the aggregated education-level specifications. Comparison of the
marginal effects of changes in poverty status to the baseline predicted probability
helps to reinforce these findings. In both models 2 and 4, the change from nonpov-
erty to poverty for individuals with a university education increases the predicted
probability of participation 2.5 times over the baseline. The predicted probability
almost doubles with the same change in poverty status for secondary school

123
Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015
124 Journal of Conflict Resolution 55(1)

Table 10. Predicted Probabilities by Education Level

Unweighted Weighted Weighted, Shia

Model 2A Model 4A Model 6A

University, P ¼ 1 0.002 (.001) 0.0007 (.001) 0.002 (.003)


University, P ¼ 0 0.0001 (.0007) 0.0002 (.0003) 0.0005 (.0007)
Secondary, P ¼ 1 0.0009 (.0007) 0.0006 (.0009) 0.002 (.003)
Secondary, P ¼ 0 0.001 (.0009) 0.0003 (.0005) 0.0008 (.001)
Preparatory, P ¼ 1 0.0003 (.0002) 0.00015 (.0002) 0.0004 (.0007)
Preparatory, P ¼ 0 0.0003 (.0003) 0.00011 (.0002) 0.0011 (.0009)
Note: Standard errors in parentheses, Poverty: P ¼ 1.

graduates. The positive effect of poverty on participation, however, shrinks and


becomes insignificant at the lower education levels. The marginal effects of poverty
are smaller for the Shia-specific regression but still positive and significant for the
university-educated subgroup. A similar pattern exists for the marginal effects of the
education-level variables, with higher levels of education having larger marginal
effects. Once again, education is only a significant positive predictor of participation
for the poor subsample. As shown in table 9, the marginal effects of the change from
not having university education to having a university degree is larger than the com-
parable change at the secondary and preparatory school levels.
Comparison of the predicted probabilities of participation across education lev-
els and poverty status, shown in table 10, reinforces several of the observations
demonstrated previously. In each specification, the poor, university-educated indi-
viduals are the most likely to participate in the terrorist organization, followed in
most cases by the poor, secondary school graduates. Comparing within each edu-
cation level, poverty increases the likelihood of participation only for individuals
with university or secondary school education. The marginal effect of poverty is
largest at higher education levels and disappears at the lower levels of schooling,
as expected based on the conditional relationship already noted. In model 4, for
example, the university educated, poor individuals are almost four times as likely
to participate in terrorist activities as university graduates who are not poor. For
secondary school graduates, those who are poor are twice as likely to participate
compared to the nonpoor subsample. Models 2 and 6 show some small differences
in this pattern, but the overall implications are similar.
This additional set of specifications shows that the relationship between poverty,
education, and participation in terrorism is sensitive to disaggregated levels of edu-
cation. The marginal effect of poverty on terrorist group membership is significant
only at the highest levels of education, but rises with small changes in educational
attainment, such as that from secondary to university level. Once again, those most
likely to participate in terrorist group activities are those who are both poor and
highly educated.

124
Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015
Kavanagh 125

Discussion and Policy Implications


Although the limitations of the data set reduce the certainty of the findings, the results
are still suggestive of a close, conditional association between education, poverty, and
participation in terrorist organizations. The conditional relationship suggests that it is
not poverty alone but poverty among more highly educated individuals that is associ-
ated with increased likelihood of terrorist group participation. Furthermore, high edu-
cation is a much stronger predictor of terrorist group involvement for poor individuals
than those who are not poor. I argued this relationship results from the rational prefer-
ence of terrorist groups for highly educated recruits combined with the low opportunity
cost of joining terrorist organization among poor, highly educated individuals.
One possible critique of the argument and findings questions the willingness of ter-
rorist groups to expend resources recruiting highly skilled individuals only to sacrifice
them in martyrdom attacks that require little technical knowledge, rather than exploiting
their potential for leadership. There are several responses to this challenge. First, martyr-
dom attacks are not always easy to carry out but require high levels of intuition and the
ability to adapt quickly to unexpected events. The use of high-skill individuals in these
operations may increase their success and effectiveness. Second, more highly educated
individuals may be more likely than their less educated peers to remain committed to
their ultimate objective and less likely to back out at the last minute. Third, if martyrdom
attacks are associated with higher status or prestige, more educated individuals may
demand such assignments in return for their membership in the organization. Research
on Palestinian organizations offers evidence for each of these explanations (Benmelech
and Berrebi 2007). Finally, the higher likelihood of participation in martyrdom opera-
tions may hold only to some educational threshold, above which individuals do take
on leadership roles. This suggestion is supported by the low number of individuals in the
data set of Hezbollah martyrs used here, who have higher than a university education.
The results suggest that poor, highly educated individuals are most likely to be
members in Hezbollah. Although the data are insufficient for a causal argument,
even the suggestive findings have several policy relevant implications. First, the
findings imply that long-term poverty may increase the likelihood of participation
in terrorism for highly skilled individuals even if it does not increase participation
in or the production of terrorism more generally. If highly educated individuals
do make more productive terrorists, then states should carefully weigh the benefits
of counterterror programs, such as sanctions and border closure that may support the
recruitment of highly skilled terrorists. Although these tactics may effectively stran-
gle some terrorist operations, they may increase the quality of attacks perpetrated by
others. Much will depend on the size of the effects. Since the marginal effects of
poverty on participation in terrorist activities appear small, governments may still
choose to use economically destructive counterterrorism where the operational gains
against terrorism are large enough. Still, more discriminating and even nonviolent
tactics that minimize economic damage or aim precisely at terrorist leadership, sup-
ply depots, or communication networks may be more effective over the long run.

125
Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015
126 Journal of Conflict Resolution 55(1)

Such a strategy might also facilitate post-conflict rebuilding and prevent resumption
of conflict after a settlement.
The theory and analysis suggest an additional economic mechanism that may effec-
tively reduce terrorism: job creation and development. The use of direct economic aid
in the form of stimulus checks given to poor, highly educated individuals in hopes of
reducing their susceptibility to radicalization is an appealing solution but also an
unsustainable, ‘‘band-aid’’ approach. This type of policy fails to address the more seri-
ous, structural reasons why highly educated individuals continue to be poor and unem-
ployed. A long-term approach would aim to reduce the likelihood of radicalization of
highly educated, poor individuals by raising their opportunity costs to participating in
terrorism. This could be done by focusing economic aid on the creation of high-skill
jobs with development potential and encouraging entrepreneurship. An effective pro-
gram would use not only economic reform and incentives but also legal reform to
remove barriers to innovation, investment, and market access. Importantly, the policy
recommendations offered here do not aim at already radicalized members or leaders
but potential recruits who may be more deterrable than hardened ideologues. By iden-
tifying ‘‘at-risk’’ populations and using policy tools to reduce the likelihood that these
individuals radicalize, this approach may have pervasive and durable effects on the
incidence of terrorism by drying up the pool of new recruits and leaders.

Future Research
The results from this analysis suggest a conditional relationship between poverty,
education, and participation in terrorism. However, these findings should be consid-
ered preliminary, due to data limitations and the use of a single case to test the
hypotheses. There are several possible extensions that would improve the certainty,
quality, and generalizability of the preliminary findings presented here. First, the
study should be repeated with improved data that identify Hezbollah members by
type of participation and include non-martyrs. Improved data would also include a
continuous economic status variable (rather than just a poor/nonpoor indicator) and
additional variables such as occupation. Second, similar analysis should be con-
ducted on other populations with significant terrorist group presence, for example,
Israel/Palestine, Sri Lanka, and even Pakistan. Such an extension would help to vali-
date the conditional relationship observed here across countries and to identify
exceptions for further investigation. It would be interesting not only to compare
results across countries but also to use pooled data to provide insight into the overall
demographics and predictors of terrorist participation on a more global scale. A third
possible extension would incorporate time-varying covariates more explicitly with
the use of time-series data on local economic conditions. Linking these data with
information on the demographic characteristics of terrorist group members over time
would provide better insight into the relationship between economic shocks and par-
ticipation in terrorist groups. These data do not exist for the Lebanese example but
could likely be obtained for other test cases. The impact of economic shocks on

126
Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015
Kavanagh 127

participation could also be assessed by comparing terrorist group participation


within two similar groups before and after an economic shock. Once again, data
limitations may prevent such a test.
Finally, the findings presented here do not disentangle the relative importance of
selection effects and opportunity cost in producing the observed higher probability
of participation by poor, highly educated individuals. Because the two mechanisms
suggest slightly different policy responses, it would be useful if we could separate
the relative importance of the two effects. This would require data on the recruitment
paths for individual militants, for example, whether the individual or the group ini-
tiated contact and whether the decision to join was one made readily or after persua-
sion. Unfortunately, this type of information may be impossible to gather.

Conclusion
This article speaks directly to the debate over the relationship between socioeconomic
status and participation in terrorism by suggesting that while poverty may not increase
participation among those with low education, it is a significant, positive predictor for
those with high education. The theory used to explain this result draws on both indi-
vidual opportunity cost and terrorist group selection patterns and implies that to under-
stand what drives individuals to participate in terrorist activities, it is not sufficient to
consider only individual attributes or motivations. Instead, analysts must also pay
attention to the demands and preferences of terrorist groups, the ability of such groups
to identify and communicate with potential members, and the opportunities that moti-
vated individuals have to access and enter terrorist organizations. In fact, the matching
process through which individuals enter a terrorist group may involve the interaction
and confluence of several determinants and may differ from case to case. Theories that
generalize predictors of terrorist group participation across cases and over time are
likely to be less useful than conditional explanations, like the one presented here, that
consider how a given determinant or characteristic affects the participation of specific
groups of individuals under various circumstances. Policy makers can exploit the
implications of such studies to develop targeted and specialized counterterror pro-
grams that prevent the radicalization of narrowly-defined, vulnerable populations.

Notes
1. The data used for the empirical analysis in this article were originally collected and coded
by Alan Krueger and Jitka Maleckova. I thank both authors for their willingness to share
their data.
2. See appendix for full description of variables.
3. The functional form can be represented as
 
PrðHezbollah ¼ 1Þ
log ¼ b0 þ b1 ðageÞ þ b2 ðhsplusÞ þ b3 ðpovertyÞ
1  PrðHezbollah ¼ 1Þ
þ b4 ðbeirutÞ þ b5 ðsouthÞ þ e:

127
Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015
128 Journal of Conflict Resolution 55(1)

The logit coefficients report the change in log odds for dependent variable that results from
a one unit change in the independent variable, all else held equal.
4. I also compute marginal effects of education level and poverty status. Marginal effects are
calculated using the first difference approach. The first difference approach compares the
probability of participation in Hezbollah at two levels of a given independent variable,
holding all other variable values at their means (or other selected value) and using the for-
mula p^1  p^0 where p^ ¼ ð1 þ exb Þ1 : The most relevant comparison for the empirical
analyses in this paper is the probability of participation when the poverty indicator is set
at ‘0’ and ’1.
5. Formally, the total effect of poverty (poverty(T)) would be equal to bpovertyðT Þ ¼
bpoverty þ bpovertyeducation  education: When education is low, this expression will be
equal to the independent effect of poverty. When education is high, the total effect of pov-
erty on participation will include both the independent and the interactive or conditional
terms. See Brambor, Clark, and Golder (2006) for more on interaction terms.

Author’s Note
The author thanks Aaron Hoffman, Mia Bloom, Rob Franzese, Jim Morrow,
Johannes Urpelainen, two anonymous reviewers, and the editor for their helpful
comments. The author also thanks Alan Krueger and Jitka Maleckova for their will-
ingness to share their data. Earlier versions of the article benefited from comments
by panel participants at the 2009 ISA Annual Convention. The data and Stata code
used for this study are available at http://jcr.sagepub.com/supplemental/. All errors
are the author’s.

Declaration of Conflicting Interests


The author declared no potential conflicts of interests with respect to the authorship
and/or publication of this article.

Funding
The author received no financial support for the research and/or authorship of this
article.

128
Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015
Kavanagh 129

Appendix

Table A1. Variable Definitions and Measurement


Age: Age in years
Ed_level: Level of schooling (primary ¼ 1, preparatory ¼ 2, secondary ¼ 3, university ¼ 4,
high studies ¼ 5)
Hezbollah: [0, 1] 0 for all nonmembers, 1 for martyrs
Poverty: [0, 1] based on eleven measures of living conditions from Population and Housing
Survey, including rooms per person, occupation, access to water and sewage, principal means
of heating, car ownership
Beirut: Dummy for residence in Beirut (1 if respondent live in Beirut)
South: Dummy for residence in the southern Lebanon (1 if respondent live in South)
hsplus: Indicator for whether the individual had more than secondary education, coded 1 if
individual has secondary education (3 on ed_level) or higher
Shia: Indicator for districts with high Shia population, coded as 1 where %Shia >69%
(there are 6)
Wt2: Weights assigned to Hezbollah sample according to their true prevalence in population
Pov  Ed: Interaction term created Poverty  Hsplus
Ed_level_pov: Interaction terms for each level of education (Higher studies  Poverty ¼
highers_pov; University  Poverty ¼ univ_pov; Secondary  Poverty ¼ Secondary 
Poverty; Preparatory  Poverty ¼ Prep_pov; Primary  Poverty ¼ primary_pov)

129
Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015
130
Table A2. Original Results, Krueger and Maleckova (2003), table 5, p. 134

Unweighted Models Weighted Models Weighted, Shia ¼ 1

1 2 3 4 5 6
*
Poverty 0.335 (.221) 0.167 (.223) 0.335 (.158) 0.167 (.162) 0.467 (.159) 0.5 (.166)**
Higher than Secondary Education 0.281 (.191) 0.171 (.193) 0.281 (.159) 0.17 (.164) 0.22 (.159) 0.279 (.167)*
** ** ** ** **
Age 0.083 (.015) 0.083 (.015) 0.083 (.008) 0.083 (.008) 0.083 (.008) 0.083 (.008)**
** **
Dummy for residence ¼ Beirut 2.199 (.219) 2.199 (.209) 0.168 (.222)**
Dummy for residence ¼ South 2.187 (.232)** 2.186 (.221)** 1.091 (.221)**
** ** ** ** **
Constant 4.886 (.365) 5.91 (.391) 5.965 (.230) 6.991 (.255) 4.658 (.232) 5.009 (1.261)**
R2 .02 .09 .02 .08 .02 .03
Observations 120,925 120,925 120,925 10,925 34,826 34,826
LR test chi-square p < .000** p < .000** p < .000** p < .000** p < .000** p < .000**

Note: Standard errors in parentheses.


* significant at .05 level.
** significant at .01 level.

Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015


Kavanagh 131

References
Angrist, Joshua. 1995. The economic returns to schooling in the West Bank and Gaza Strip.
American Economic Review 85 (5):1065-87.
Atran, Scott. 2006. The moral logic and growth of suicide terrorism. Washington Quarterly 29
(2):127-47.
Benmelech, Efraim, and Claude Berrebi. 2007. Human capital and productivity of suicide
bombers. Journal of Economic Perspectives 21 (3):223-38.
Berman, Eli, and David Laitin. 2006. Hard targets: Theory and evidence on suicide attacks.
CISAC working paper. http://dss.ucsd.edu/~elib/Hardtargets.pdf (accessed 6/10/2010).
Berrebi, Claude. 2007. Evidence about the link between education, poverty and terrorism
among Palestinians. Peace Economics, Peace Science, and Public Policy 13(1): 1–36.
Blomberg, S. Brock, and Gregory Hess. 2008a. From (no) butter to guns?: Understanding the
economic role in transnational terrorism. In Terrorism, economic development, and polit-
ical openness, ed. Keefer Philip and Norman Loayza, 83-115. New York: Cambridge
University Press.
Blomberg, S. Brock, and Gregory Hess. 2008b. The Lexus and the olive branch: Globaliza-
tion, democratization, and terrorism. In Terrorism, economic development, and political
openness, ed. Keefer Philip and Norman Loayza, 116-47. New York: Cambridge
University Press.
Blomberg, S. Brock, Gregory Hess, and Akila Weerapana. 2004. An economic model of ter-
rorism. Conflict Management and Peace Science 21 (1):17-28.
Brambor, Thomas, William Clark, and Matt Golder. 2006. Understanding interaction models:
Improving empirical analyses. Political Analysis 14 (1):63-82.
Brym, Robert, and Bader Araj. 2006. Suicide bombing as strategy and interaction: The case of
the Second Intifada. Social Forces 84 (4):1969-86.
Bueno de Mesquita, Ethan. 2005. Quality of terror. American Journal of Political Science 49
(3):515-30.
Burgoon, Brian. 2006. On welfare and terror: Social welfare policies and political-economic
roots of terrorism. Journal of Conflict Resolution 50 (2):176-203.
Crenshaw, Martha. 2001. Counterterrorism and the political process. Journal of Conflict Res-
olution 24 (5):329-37.
Eken, Sena, Paul Cashin, S. Nuri Erbas, Jose Martelinao, and Adana Mazarei. 1995. Eco-
nomic Dislocation and Recovery in Lebanon. Occasional paper 20, Washington, DC.
International Monetary Fund.
Enders, Walter, and Todd Sandler. 2000. Is transnational terrorism becoming more threaten-
ing?: A time series investigation. Journal of Conflict Resolution 44 (3):307-32.
Enders, Walter, and Todd Sandler. 2006. Distribution of transnational terrorism among coun-
tries by income class and geography after 9/11. International Studies Quarterly 50 (20):
367-93.
Gujarati, Damodar. 2004. Basic Econometrics, 4th Edition. New York: McGraw Hill.
Hassan, Nasra. 2001. Talking to human bombs. New Yorker 77 (36):36-42.
Homer-Dixon, Thomas. 2002. The rise of complex terrorism. Foreign Policy 128:52-63.

131
Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015
132 Journal of Conflict Resolution 55(1)

Honaker, James. 2008. Unemployment and violence in Northern Ireland: A missing data
model for ecological inference. UCLA working paper. http://tercer.bol.ucla.edu/papers/
ni.pdf (accessed 6/8/2010).
Hudson, Rex. 1999. The sociology and psychology of terrorism: Who becomes a terrorist and
why? Washington, DC: Federal Research Division.
Knudsen, Are. 2005. Precarious peacebuilding: Post-war Lebanon, 1990-2005. CMI working
paper, Chr. Michelsen Institute, Bergen.
Krueger, Alan. 2007. What makes a terrorist?: Economics and the roots of terrorism.
Princeton: Princeton University Press.
Krueger, Alan, and David Laitin. 2008. Kto kogo?: A cross country study of the origins and
targets of terrorism. In Terrorism, economic development, and political openness, ed. Kee-
fer Philip and Norman Loayza, 148-73. New York: Cambridge University Press.
Krueger, Alan, and Jitka Maleckova. 2003. Education, poverty, and terrorism: Is there is a
causal connection? Journal of Economic Perspectives 17 (4):119-44.
Lichbach, Mark Irving. 1987. Deterrence or escalation: The puzzle of aggregate studies of
repression and dissent. Journal of Conflict Resolution 31 (2):266-97.
Manski, Charles, and Steven Lerner. 1977. The estimation of choice probabilities from choice
base samples. Econometrica 45 (8):1977-88.
Norton, Augustus Robert. 2009. Hezbollah: A short history. Princeton: Princeton University
Press.
Sageman, Marc. 2004. Understanding terror networks. Philadelphia: University of
Pennsylvania Press.
Sprinzak, Ehud. 1990. The psychopolitical formation of extreme left terrorism. In Origins of
terrorism: Psychologies, ideologies, theologies, states of mind, ed. Reich Walter and
Walter Laqueur, 65-85. Princeton: Woodrow Wilson Center Press.
White, Robert. 1989. From peaceful protest to guerrilla war: Micromobilization of the Provi-
sional Irish Republican Army. American Journal of Sociology 94 (6):1277-302.
White, Robert. 1993. On measuring political violence: Northern Ireland 1969 to 1980.
American Sociological Review 58:575-85.

132
Downloaded from jcr.sagepub.com at UNIV OF ILLINOIS URBANA on March 10, 2015

You might also like